Skip to slide
33 / 53

CH 04 · DeepSeek-R1, Learning to Reason Through Reinforcement Learning · 07/07

Read as one page

The one-sentence takeaway

DeepSeek-R1 showed that a model can teach itself to reason, growing longer chains of thought and self-checking habits on its own, when it is trained with reinforcement learning and rewarded simply for reaching verifiably correct answers, with a light touch of clean example data added only to tidy up how it presents its thinking.

Next: Chapter 5, Recursive Language Models, where a model learns to handle inputs far larger than its own memory.

Finished this idea?

Your learning progress stays in this browser.